Machine learning prediction and calibration of cellulose-based solid-phase extraction performance for pharmaceuticals...

Area of Science:

• Environmental Chemistry
• Analytical Chemistry
• Chemometrics

Background:

• Cellulose-based solid-phase extraction (SPE) is promising for concentrating trace pharmaceuticals in complex water samples.
• Cross-laboratory transfer of these methods is hindered by variations in experimental design and matrix effects.

Purpose of the Study:

• To evaluate the predictive performance of machine learning models for cellulose-based SPE of trace pharmaceuticals.
• To identify key method and matrix descriptors influencing extraction efficiency and detection limits.
• To provide guidance for defensible local validation and method transfer.

Main Methods:

• A systematic review and meta-analysis of 637 experiments from 36 reports (2015-2025).
• Modeling using ElasticNet (EN), XGBoost (XGB), and Random Forest Regressor (RFR) with 29 method/matrix descriptors.
• Nested cross-validation with conformal prediction to estimate out-of-study performance and confidence intervals for recovery, matrix recovery ratio (MRR), enrichment factor (EF), limit of detection (LOD), and limit of quantification (LOQ).

Main Results:

• ElasticNet (EN) demonstrated superior prediction for sensitivity endpoints: EF (R²=0.99999), LOD (R²=0.99985), and LOQ (R²=0.99914).
• Random Forest (RF) showed the strongest correlation for recovery and MRR but remained weakly predictive (R² ≈ -0.52 and -1.03, respectively).
• Prediction performance varied significantly with matrix type, particularly for wastewater and river samples.

Conclusions:

• Machine learning, especially ElasticNet, can accurately predict the performance of cellulose-based SPE for trace pharmaceutical analysis.
• Method and matrix descriptors are crucial for reliable prediction and successful method transfer.
• Decision maps derived from model contrasts offer practical guidance for optimizing SPE protocols and validation.